17079: The Economics of Mobiles and Mainframe - SHARE · 17079: The Economics of Mobiles and Mainframe ... Installed smart devices probably now exceed PCs ... and mobile applications

Insert

Custom

Session

QR if

Desired.

17079: The Economics of Mobiles

and Mainframe

David Rhoderick, IBM

© 2015 IBM Corporation

IBM Competitive Project Office

Agenda

• Mobile revolution is trending more than “eBusiness 1.0”

• MobileFirst deployment choices

• Mobile Workload Pricing

• Architectural choices for data serving

The economics of mobile and the mainframe4



Mobile business is a much bigger phenomenon than “ebusiness 1.0”

5

1.2B(NOTE: World population

is about 7.1B)

Time Magazine, January 2014Projections of mobile growth and PC decline based on Gartner data

The economics of mobile and the mainframe



Installed smart devices probably now exceed PCs (trailing 5y PC sales and trailing 24m phone/tablet sales)

6 The economics of mobile and the mainframe



Other signs that the technology is becoming widely adopted

7

Sources: Javelin Strategy and Research, Alix PartnersSee http://online.wsj.com/news/interactive/mobile0326A?ref=SB10001424052702303847804579481811781070456


http://online.wsj.com/news/interactive/mobile0326A?ref=SB10001424052702303847804579481811781070456



8 The economics of mobile and the mainframeSource: Mary Meeker, 5/28/2014



9 The economics of mobile and the mainframeThe economics of mobile and the mainframe



Web-enablement of back-end functions is effective way to grow business – “ebusiness” includes modernizing existing assets

In September 2012, the independent research firm Vanson Bourne studied IT trends and

interviewed 520 CIOs from large enterprises across a range of industries in the U.S.,

Europe & Asia

62% have enabled back-end

functions to support e-commerce

and mobile applications

– Mature technology and tooling

helps do this quickly and easily

68% also believe mobile computing

will drive MIPS growth




11

System z installed capacity grown at 17% CAGR – partly due to “eBusiness 1.0” – direct mobile access will accelerate this


* these MIPS could also be used for z/VM, z/VSE or z/TPF Operating Systems



There are proactive – and reactive – reasons for opening a mobile channel

12

“… if we don’t offer the functionality desired by our customers, there’s

significant risk of falling behind in the market where competitors are

already offering these services and we will fail to attract sufficient custom to achieve the target to grow market

share“… Smartphone channels offer a true alternative in many instances to the

internet self-service channels…”

Sources: Javelin Strategy and Research, Alix PartnersSee http://online.wsj.com/news/interactive/mobile0326A?ref=SB10001424052702303847804579481811781070456


“a noticeable dip in customers

contacting call centres”

http://online.wsj.com/news/interactive/mobile0326A?ref=SB10001424052702303847804579481811781070456



Customer

Teller

ATM

Customer Service Manager

Online / Mobile

@

When a customer's card is not returned by the ATM then…

When a pending transaction will cause an account to be overdrawn then…

Personal

Promote overdraft protection offer

Send SMS with location of nearest branch

When a large payment has been requested then…

Request biometric authentication

Request biometric authentication

Mobile is changing the way customers interact – far more possibilities




WOLA

WOLA

IBM z/OS Connect mobile enables backend on System z

WOLA Link Server Task

CICSPrograms

IMS dependent regions

BlueMix

IMS

VSAM

DB2

DLI

zosConnectJSON to/from

byte[] (Cobol

copybook)

CICS

WOLA over OTMA

WOLA direct

BatchzosConnectJSON to/from

byte[] (C

structure)

zosConnectJSON to/from

byte[] (PL/I

structure)

WOLA

WAS Liberty z/OS

RESTful

RESTful

RESTful

z/OS ConnectA service that encapsulate calling z/OS target applications using REST calls. z/OS Connect will support JSON payloads for calls from external cloud or mobile-based clients and will enable the conversion of the payload to the target program's expected format. It will also provide the response payload conversion from a byte array into JSON format before returning the response to the caller.

Data binding conversion/routing

Mobile/APIM

Security / Auditing /

Metering

CloudOEJavascript

CloudOE Java




Centralized server technology provides a platform to manage and drive all mobile applications


z/VM

Linux

MobileFirstFoundation

IBM MobileFirst

Windows

Android

Blackberry

Apple

z/OS

WAS or MQ

z/OS

DB2

z/OS

CICS or IMS

JSON

Simpler to maintain

Secure

Standards-based

Easier and faster to develop apps

Mobile protocol connectivity

with core zEnterprise applications

including CICS, IMS, TPF, MQ, WMB and DB2



DevOps capabilities for mobile applications with enterprise systems

• Quickly design, code, build, test, and deploy mobile apps that run on a wide variety of mobile platforms

• Refactor and extend existing back-end services to provide an optimal mobile experience

• Stub out back-ends to simplify test and development

• Automate testing for native and hybrid mobile apps

Accelerate mobile application development

Enabling you to…

Deploy

Code

Test

Open Lifecycle and Service ManagementIntegration Platform

Continuous Delivery

Construct, debug, and test mobileand Web applications

Design

Refactor and extend existing logic as mobile-consumable services

Rational Test WorkbenchIBM MobileFirst Platform

IBM DevOps services for BluemixRational Application Developer

IBM UrbanCode DeployRational Developer for the Enterprise

CICS Interdependency AnalyzerCICS Performance Analyzer




18

University of Florida goes mobile

Data provided to students real timeMobile formatted information of class schedules, textbooks, academic dates, grades, emergency information and campus map

IBM SolutionAccessing CICS with System z information via smartphones

Enabling 50,000 students, 5,400 faculty members and staff access to online features anytime, anywhere

Up to 1M transactions/day




“Mobile banking is a fast growing channel for Halkbank. As for all other channels, we run the transactions themselves on our CICS and DB2 for z/OS systems, which helps to ensure the highest levels of performance, availability and security”

“Core enterprise applications process transactions — lots of transactions — and so can’t be scaled by adding load balancers and spawning new copies of services. They scale by adding capacity to the transaction processing engine”

25 years of continuous operationOn CICS®, DB2®, and System z®

Ayhan Yalkut, System z Manager, Halkbank.

The Turkish state pays most of the government salaries through the bank, causing a once-monthly peak in daily transactions, increasing from 25m to 50m

Scalable

“In more than 25 years of continuous operations, we have never had any unplanned downtime”

Reliable

Mobile




IBM studied MobileFirst Platform running on Linux/x86 versus Linux/z

CICS/DB2 Application on System of Records (GENAPP)– Montpellier Benchmarking System Center

Tested 400 concurrent threads with 60% Login, 30% Add or

Delete and 10% Update

Compared Throughput, Response Time and TCA per TPS

between Linux on System z versus x86– 61% higher throughput

– 36% lower response time

– 10% lower TCA




Architecture of MobileFirst Platform Performance Test

z/OS*Linux on z* or x86

WAS ND 8.5.5 IBM HTTP Server 8.5

MobileFirst Platform 6.1 DB2 ESE 10.1

CICS V5R1 (GENAPP)

WAS ND 8.5.5 MobileFirstPlatform 6.1

Mobile Users

DB2 10 for z/OSWAS ND 8.5.5

MobileFirstPlatform 6.1

WAS ND 8.5.5 MobileFirstPlatform 6.1

* Estimated performance, sizing and cost for z13 based on tests conducted on zEC12

This is based on an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. Test involved measuring throughput in transactions per second and response time for executing a materially identical mobile transaction processing workload in a controlled laboratory environment with comparable tuning and sizing. Prices, where applicable, are based on published US list prices as of 12/31/2014 for both IBM and competitor. Price comparison based on 3 Year Total Cost of Acquisition (TCA) includes all HW, SW and 3 years of service & support. Sizing shown is for Production to which 30% is added for System z for Dev/QA and CBU pricing for DR and 2x for Distributed.




Raw test results summary

Worklight server on z/Linux Worklight server on x86

Number of CPU

per Worklight

server

Number of

users

Response

Time (ms)

Throughput

(TPS)

Max number of

physical CPU used

(on 8CPs)

Response

Time (ms)

Throughput

(TPS)

Max number of

physical CPU used

(on 16Cores)

1 10 42.9 295.71 0.56 (avg 0.55) 50.2 242.89 0.75 (avg0.68)

30 44.5 850.4 2.17 (avg 1.90) 54.9 652.6 1.7 (avg1.7)

50 47.7 1304.1 2.32 (avg 2.21) 59.5 991.8 2.5 (avg2.14)

100 60.3 2000 3.30 (avg 3.18) 75.4 1518.1 3.87 (avg3.87)

200 104.4 2195.8 3.67 (avg 3.41) 132.2 1676.2 4.26 (avg4.26)

400 189.9 2361.2 3.88 (avg 3.78) 256.8 1699.1 4.64 (avg4.64)

600 293.1 2248.1 3.96 (avg 3.71) 357.8 1843.8 4.7 (avg4.7)

2 10 42.8 296.34 0.61 (avg 0.59) 50 244.12 0.75 (avg0.75)

30 43 881 2.3 (avg 1.74) 54.4 659 1.83 (avg1.83)

50 44.4 1418.9 2.48 (avg 2.42) 62.5 931.8 2.3 (avg2.3)

100 48.6 2554.3 4.30 (avg 4.21) 68.7 1675 4.14 (avg4.14)

200 70.8 3320.2 6.02 (avg 5.34) 95.7 2337.7 6.08 (avg6.08)

400 131.4 3446.1 5.96 (avg 5.53) 205.4 2145.2 4.96 (avg4.96)

600 215.1 3126.9 5.97 (avg 5.08) 291.6 2264.7 6.05 (avg6.05)

• Test 4: Mix (60% Login, 30% Add or Delete, 10% Update)




MobileFirst Platform scales better on Linux on System z than x86

0

500

1000

1500

2000

2500

3000

3500

4000

0 100 200 300 400 500

Throughput versus Concurrent Users

Linux on System z

x86

Thro

ugh

put

(TP

S)



This is based on an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. Test involved measuring throughput in transactions per second and response time for executing a materially identical mobile transaction processing workload in a controlled laboratory environment with comparable tuning and sizing. Prices, where applicable, are based on US prices as of 12/31/2014 for both IBM and competitor. Price comparison based on 3 Year Total Cost of Acquisition (TCA) includes all HW, SW and 3 years of service & support. Sizing shown is for Production to which 30% is added for System z for Dev/QA and CBU pricing for DR and 2x for Distributed.

Concurrent users



MobileFirst Platform provides lower Response Time on Linux on z than x86

0

50

100

150

200

250

0 100 200 300 400 500

Response Time versus Concurrent Users

Linux on System z

x86

Concurrent users

Respon

se

Tim

e (

ms)






MobileFirst Platform is expected to provide lower front-end cost on Linux on z* than x86, for > 30 concurrent users


$0

$500

$1,000

$1,500

$2,000

$2,500

$3,000

0 100 200 300 400 500

TCA per TPS versus Concurrent Users

Linux on System z

x86

Concurrent users

TC

A p

er

TP

S (

$)



Tesco planned on 50,000 concurrent

users for Black Friday – and got

175,000!



Competitor x86 System

Intel E5-2697v2 2.7GHz 6co

Which platform provides the lowest TCA over 3 years?

400 concurrent connections

60% Login, 30% Add or Delete,

10% Update

3,446 TPS

131ms RT

$439 per TPS(3 yr. TCA, SoE)

Prod + Dev/QA + DR

2,145 TPS

205ms RT

$693 per TPS(3 yr. TCA, SoE)

Prod + Dev/QA + DR

Mobile Users

MobileFirst

Platform

WAS ND

DB2

x86 VM

CICS

DB2

z/OS

Mobile Insurance workload

27

MobileFirst

Platform

WAS ND

DB2

z/VM

CICS

DB2

z/OS

z13-409 + 5 IFLs

z13-405

Running MobileFirst Platform on z13* vs x86 with similar z Systems* System of Record should increase throughput by 61% and reduce response time by 36% and front-end cost by 37%



Estimated

37% lower cost

for systems compared




Why does Linux on z perform better than Intel?

Better hypervisor– z/VM overhead of 5-8% vs VMware overhead of 20-25%

Co-location of Worklight server with the data– Hipersockets on z reduces response time

For a 50 user run in both architecture we get a network IO write of 966MB/s on zLinux

versus 651MB/s on Intel (so 49% faster with hipersocket) and a network IO read of

1180MB/s on zLinux versus 820MB/s on Intel (so 45% faster with hipersocket)

– Encrypted connection between Worklight server with backend may not

be required, resulting in less CPU overhead

SMT2 enables 30% improvement in IFL throughput

Better caching architecture in System z processors– We didn’t measure this but expect that the multi-tiered processor

cache is a performance booster




Mobile customer requirements are more demanding – fast response and data accuracy are important

29

“Millenials” are potential new customers –

with little patience –expect the data to be instant and up-

to-date

#1 reason people in the UK switch banks is dissatisfaction

with the old bank’s app




30

http://www.bbc.com/news/business-26387754

Immediate access to the “System of Record” is now expected

5,500 customers logging on per minute caused slowdowns

Consumers were checking salary/benefit payments

Some had “sticky fingers”


http://www.bbc.com/news/business-26387754



IBM studied price-performance of read-only caching

“Pulling” SOR data into a mid-tier cache shields mainframe from query-only (“non-

revenue producing”) traffic• For read-only mobile workloads – mobile-initiated writes must be processed on

System Of Record

• WXS used for technology study

• Use-case may be sub-optimal (mobile bank users want to see update data, not old)

Tested “Sticky Finger” scenario with 500 users, each doing 20 reads per login

session with 100ms think time and a 1s cache invalidation• Assume 500 users with 100ms think time equivalent to 5,000 users and 1s think time

Quantified TCA savings for “Sticky Finger” scenario reading CICS/DB2 data with

WebSphere eXtreme Scale (WXS) on zLinux as front-end cache• 64% savings for WXS caching vs business as usual (BAU)

• 29% savings with MWP

Quantified TCA increase for “Blended” scenario with 500 users (70% do 1

read/session, 25% do 4 reads/session and 5% do 20 reads/session)• Just 4% savings for WXS caching vs BAU

• With new MWP, no caching (BAU) is 45% less cost




Architecture of read-only mobile workload with WXS caching


z/OSLinux on z

Rest Client Application (WAS Liberty)

CICS V5R1

Rest Server Application (Liberty)

WebSphereExtreme

Scale

Mobile Users

DB2 11 for z/OS

Only used when caching turned on(i.e. lower configurations on

subsequent charts)



WXS caching reduces TCA for sticky finger with think-time usermobile workloads by 64%




20 reads/session with 100ms

think time (forcing cache refresh)

1 second cache invalidation (WXS

scenario)

$21.8M (3 yr. TCA)

Prod

(BAU)

$7.9M (3 yr. TCA)

Prod

Mobile Users

Liberty

WXS

z/VM

CICS

DB2

z/OS

Mobile workload drives minimum throughput of 5,200 transactions per second and response time of 5ms

64%

lower cost!

Liberty

z/VM

CICS

DB2

z/OS

zEC12-705 + 3 IFLs

zEC12-407 + 4 IFLs



IBM announced Mobile Workload Pricing for z/OS …

An enhancement to sub-capacity pricing

Normalizes the rate of transaction growth

Mitigates the impact of Mobile on MLC charges where higher

transaction volumes cause a spike in machine utilization

Works like an MSU “off-load” from a software pricing perspective,

similar to Integrated Workload Pricing, not a defined price discount

No infrastructure changes required (i.e. no separate LPARs) … rather an

enhanced way of reporting sub-capacity MSUs

Available to all enterprises running an zEC12 or zBC12 that meet the

Mobile workload tracking requirements

Formal announcement letter available now!

Mobile Workload Pricing for z/OS helps alleviate spikesImproves the cost of growth for mobile transactions processed in System z environments such as CICS, IMS and DB2




Mobile Workload Pricing for z/OS Reporting Process

New sub-capacity reporting tool (June 2014)

A new Windows-based tool is being developed to report sub-capacity MSUs and make adjustments to

LPAR peaks based on Mobile transaction data

– Standard SCRT methodology plus new feature to adjust for Mobile workload impact

– New tool will replace SCRT for customers who take advantage of Mobile Workload Pricing

Customer requirements – Important!

Tag and track Mobile transactions and produce a file showing Mobile CPU consumption each month

– Track Mobile transactions, including CPU seconds, on an hourly basis per LPAR

– Load the resulting data file into the reporting tool each month (IBM-specified CSV format)

– Run the IBM-supplied tool and submit the results to IBM each month (Replaces SCRT process)

Monthly peak calculation for billing

The tool will subtract 60% of the measured Mobile MSUs (customer input) from a given LPAR in each

hour, adjusting the total LPAR MSU value for that hour

– When an LPAR value is adjusted, all software running in the LPAR will benefit from lower MSUs

– Tool will calculate the monthly MSU peak for a given machine using the adjusted MSU values

– From a software pricing perspective this will look like a partial “off-load”

Provides benefit when Mobile workloads contribute to monthly peak MSUs;

Off-peak MSU adjustments will not affect MSUs used for billing by definition




z/OS &

Other

CICS

Other

1,380 MSUs

CICS,

z/OS &

Other

100

80

Adjusted LPAR peak for month Pricing & billing BAU based on peak

1,500 MSUs

Measure LPAR peak

with standard methodology

(SMF70)

1

Measure CICS usage

Capture SMF 89 record in

new IBM reporting tool

(replaces SCRT)

2

Capture CICS transaction

details and filter by

transaction type, mobile or not

3

Subtract 60% of mobile usage:

-60% * 2004

Adjust LPAR peak

with new reporting tool5

6

Mobile

* Figures are for illustrative purposes only. Tracking process and records will vary by customer

Other

Mobile200

100

Transaction

definitions

Mobile Workload Pricing for z/OS helps mitigate spikes caused by increased mobile usage – and lowers costs


CICS

300

CICS

Usage



Consider a blend of different use cases

70% are regular users (1 read/session)

– For these cases, the cache is additional overhead and cost

25% are frequent-read users (4 reads/session)

– Cache shields mainframe from 3 reads

– No cache-refresh

5% do 20 reads/session

– “Sticky finger” with cache-refresh

– 2 mainframe reads




WXS caching reduces TCA for blended mobile workloads by 4% (without MWP)




70% do 1 read/session;

25% do 4 reads/session;

5% do 20 reads/session with

100ms think time

1 second cache invalidation

(WXS scenario)

$19.8M (3 yr. TCA)

Prod

$19M (3 yr. TCA)

Prod

Mobile Users

Mobile workload drives minimum throughput of 6,300 transactions per second and response time of 12ms 4%

lower cost!

Liberty

z/VM

CICS

DB2

z/OS

zEC12-704 + 5 IFLs

635 MSUs

zEC12-704+ 5 IFLs

529 MSUs

Liberty

WXS

z/VM

CICS

DB2

z/OS



Mobile Workload Pricing provides 45% lower TCA than WXS Caching for blended mobile workloads







100ms think time


(WXS scenario)

$10.4M (3 yr. TCA)

Prod Mobile Workload Pricing

$19M (3 yr. TCA)

Prod

Mobile Users

Mobile read-only workload driving minimum throughput of 6,300 transactions per second and response time of 12ms

Liberty

z/VM

CICS

DB2

z/OS

zEC12-704 + 5 IFLs

635 MSUs

zEC12-704+ 5 IFLs

529 MSUs

Liberty

WXS

z/VM

CICS

DB2

z/OS

45%

lower cost!



Where best to serve SOR data to mobile devices?

1. Intermediate tier using Extract Transform Load (ETL) processing– Daily refresh

2. Intermediate tier using ETL with “trickle feed”– Intra-day updates

3. Intermediate tier using on-demand caching (IBM studied this)– Read-only cache (no writing BACK to the mainframe)

– Oracle Coherence, WebSphere eXtremeScale …

4. Feed directly from the mainframe source SOR data– Leverage new Mobile Workload Pricing

5. Use “Push” technology– See Cuomo/Vila IBM Redbooks Point-of-View: Mobile Design Patterns: Push, Don't Pull

Evaluation criteria:– Provide enhanced consumer experience (data freshness, responsiveness)

– Deploy with enterprise QoS (RAS, security, scalability) at lowest cost

Also see Williams/van der Wal Redbooks Point-of-View: System z in a Mobile World


http://www.redbooks.ibm.com/abstracts/redp5072.html?Open

http://www.redbooks.ibm.com/abstracts/redp5088.html?Open



OperationalData

A large Asian bank:

One mainframe devoted exclusively to bulk data transfers

ETL consuming 8% of total distributed core and 18% of total MIPS

A large European bank:

120 database images created from bulk data transfers

1,000 applications on 750 cores with 14,000 software titles

ETL consuming 28% of total distributed cores and 16% of total MIPS

AnalyticalData

AnalyticalData

AnalyticalData

AnalyticalData

AnalyticalData

ETL rule of thumb: only 20% of the data is ever used

Actual studies confirm ETL data has significant unintended costs (and on average, 12 hour-old data)




Significant costs (often hidden) are involved when moving data off the mainframe

Operational

Data

Analytical

Data

Analytical

Data

Analytical

Data

Analytical

Data

Estimated 4 yr. cost summary

System costs =

$9.86M

Labor costs =

$0.39M

Total = $10.3M

z13

x86

Assuming 4 cores on z13 running at 85% utilization and 12

cores on x86 servers run at 15% utilization, transfer will burn

519 MIPS and use 30 x86 cores per dayThis is based on an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. Test involved measuring in a controlled laboratory environment elapsed time for system and

administrator to extract, send and receive 130GB file from z13 to an x86 server running with 12 x Xeon 2.4GHz E5-2440 processors. Prices, where applicable, are based on US prices as of 12/31/2014 for both IBM

and competitor. Estimated amortized cost from 4 Year Total Cost of Acquisition (TCA) that includes all HW, SW (OS, DB and tools) and 4 years of service & support. For Labor costs, used annual burdened rate of

$159,600 for IT Administrator for z Systems and x86. Results may not be typical and will vary based on actual workload, configuration, applications, queries and other variables in a production environment. Users of

this document should verify the applicable data for their specific environment. The economics of mobile and the mainframe42



Replicated copy is unacceptable – the data is always old –resulting in excessive complexity and cost with negative ROI

Load Balance

And

Fail

Over

APS/

ECPRCDF

ETL

SQLSQL

Channels

Systems of Record

Batch Extracts

Change Notifications

Transaction Authorizations

and

Transaction Updates

APS/

ECPRCDF

ETL

Business

Services

SQLSQL

Channels

Batch Extracts

Change Notifications

Transaction Authorizations

And

Transaction Updates

Minneapolis

Tempe

SQL Failover

Tempe APS/ECPR

to NOC CDF

SQL Failover

NOC APS/ECPR

to Tempe CDF

Business

Services


2,000+ jobs

each night

transfer files to

the ETL server

Autosys

detects files

are ready

Extract files

created each

night

Ab Initio reads

files and

processes

updates

warehouse



Will you reduce cost by moving your Mobile data off System z ?IBM studied* intermediate tier using ETL with “trickle feed”

CICS

DB2

Pre-integrated

Competitor

App Server

Eighth Unit

Pre-integrated

Competitor

DB Eighth

Unit

We compared the cost of this

vs this

Add MobileFirst Platform to create a

new mobile channel to existing data

and transactions.

Move SoR data from System z to a new SoR off-platform, and serve

mobile transactions from there.* Estimated performance, sizing and cost for z13 based on tests conducted on zEC12

This is based on an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. Test involved measuring throughput in transactions per second and response time for executing a materially identical mobile transaction processing workload in a controlled laboratory environment with comparable tuning and sizing. Prices, where applicable, are based on published US list prices as of 12/31/2014 for both IBM and competitor. Price comparison based on 3 Year Total Cost of Acquisition (TCA) includes all HW, SW and 3 years of service & support. Sizing shown is for Production to which 30% is added for System z for Dev/QA and CBU pricing for DR and 2x for Distributed.




Which platform provides the lowest TCA over 3 years? $11.2M (3 yr. TCA)

Prod + Dev/QA + DR

Mobile Workload Pricing

$18.6M (3 yr. TCA)

Prod + Dev/QA + DR

CICS

DB2

z/OSMobile read-only workload driving minimum

throughput of 6,300 transaction persecond and response time of 12ms

Replicating z Systems System of Record for Mobile Workloads To Competitive System is expected to increase TCA by 66% versus co-locating MobileFirstPlatform and using Mobile Workload Pricing

MobileFirst

Platform

WAS ND

DB2

z/VM

CICS

DB2

z/OS

z13-606 + 4 IFLs*

z13-606 Pre-integrated

Competitor*

Eighth Unit

(30 cores pro-rated)

Competitor*

App +

Caching

Software

VM





100ms think time


Mobile Users

* Competitor Caching and Database sizing estimated from WebSphere Extreme Scale Caching Test. ** Estimated performance, sizing and cost for z13 based on tests conducted on zEC12

This is based on an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. Test involved executing a materially identical mobile transaction processing workload in a controlled laboratory environment with comparable tuning and sizing. Prices, where applicable, are based on US prices as of 12/31/2014 for both IBM and competitor. Price comparison based on 3 Year Total Cost of Acquisition (TCA) includes all HW, SW and 3 years of service & support. Sizing shown is for Production to which 30% is added for System z for Dev/QA and CBU pricing for DR and 2x for Distributed.

Competitor*

Database

Linux

Pre-integrated

Competitor*

Eighth Unit

(24 cores)

Estimated

66% higher cost

for systems compared




PUSH versus PULL study

Cost case simulates an Automated Benefits Entitlement Deposit scenario

that happens midnight end of the month

Assumes 5% of customers are “sticky finger” users– 500 concurrent connections (around 5,000 TPS)

When a deposit arrives end of the month for all 10K customers over a

second of time, the throughput supported by the deposit transaction is

10,000 TPS

– Need 2x hardware used for “sticky finger” scenario

– To process a PUSH transaction for all 10K customers using CICS

Event Processing (CEP) CPU overhead is 14%




CPU cost of CICS Event Processing


Capturing rate (per task): T1: 1 in 100, T2: 1 in 10, T3: 1 in 4, T4: 1 in 2, T5: 1 in 1



PUSH reduces TCA by 51% over PULL assuming 5% of customers with “sticky finger” behavior


$11.2M (3 yr. TCA)

Prod (Using Mobile Workload Pricing)

$5.5M (3 yr. TCA)

Prod

MQ

CICS

DB2

z/OS

51%

lower cost!

Liberty

z/VM

CICS

DB2

z/OS 500 concurrent connections (5% of

customers)

20 reads/session with 100ms

think time (forcing a cache refresh)

1 second cache invalidation (WXS

scenario)

Mobile Users

Liberty

z/VM

10,000 PUSH events per second for

all customers

Mobile Users

PULL

PUSH



PUSH reduces TCA by 73% over PULL for blended mobile workloads


$10.4M (3 yr. TCA)

Prod (Using Mobile Workload Pricing)

$2.8M (3 yr. TCA)

Prod

MQ

CICS

DB2

z/OS

73%

lower cost!

Liberty

z/VM

CICS

DB2

z/OS

Mobile Users

Liberty

z/VM

5,000 PUSH events per second for

all customers

Mobile Users

PULL

PUSH





100ms think time


(WXS scenario)



Summary


Mobile revolution is trending much more than “eBusiness 1.0”Anticipate huge volumes of transactions, with unpredictable spikesNew opportunities, new customers

MobileFirst for Linux on zExtending mainframe workloads to mobile devices is cost-effective and easyPlus security, latency, DR advantages…

Mobile Workload PricingNew challenges – usage demands are tougher, less tolerantNew mobile workloads drive monthly peaksSignificant benefit, but initial version requires some record-keeping

Architectural choices for data servingNo need for an intermediate DB or cache – inefficient, costly, stalePushing events probably better than pulling data




Questions?

Documents

17079: The Economics of Mobiles and Mainframe - SHARE · 17079: The Economics of Mobiles and Mainframe ... Installed smart devices probably now exceed PCs ... and mobile applications